Top News

ChatGPT: Optimizing Language Models for Dialogue

OpenAI has launched a new powerful AI-powered language model, ChatGPT, which is trained to interact in a conversational manner. The model is capable of answering follow-up questions, admitting its mistakes, challenging incorrect premises, and rejecting inappropriate requests. It has been trained using Reinforcement Learning from Human Feedback, a method that utilizes human feedback to fine-tune AI models. ChatGPT is a sibling model to InstructGPT, which is trained to follow instructions provided in a prompt and provide a detailed response. The model is fine-tuned from GPT-3.5.

Our take: This may be one of the few official AI release blog posts that really understates the potential impact and applications of what is being released. Within a couple of days, people found a myriad of impressive capabilities of ChatGPT: finding bugs in code and explaining them, finding exploits in code, writing genuinely good song lyrics, writing essays and giving detailed feedback on them, explaining concepts in the voice of fictional characters, acting as a virtual machine, browsing the web to answer questions, and more. You can try out the model today at https://chat.openai.com/chat.

The model is far from perfect (for example, here it provides a convincing explanation of regex, a computer science concept, but the explanation is actually wrong) and still suffers similar limitations as previous language models (hard to make guarantees, hard to perform precise reasoning, hard to avoid “hallucinating” incorrect information). However, ChatGPT still marks a significant improvement in the utility of large language models across a wide variety of tasks, and no doubt OpenAI is using the user feedback on ChatGPT to improve the training of the rumored GPT-4. It is a very exciting time in the world of generative AI, and we will likely see more significant developments in just the next few months.

The DOJ is reportedly investigating rent-setting software company RealPage

An image showing the Department of Justice lettering

The Department of Justice (DOJ) is reportedly investigating RealPage, a US-based property technology company, to determine if its YieldStar software allows landlords to coordinate and raise rent across the country. The investigation comes after ProPublica published an article that claimed RealPage’s YieldStar algorithm uses non-public rental rates gathered from landlords and property managers to make rental rate recommendations to its clients, indirectly giving landlords access to their competitors’ pricing. If true, this would potentially violate federal law.

Our take: This is definitely a warning for those who are trying to deploy AI in ways that can lead to potential harm. It's clear that using algorithms to drive up rental prices is not only unfair to tenants, but it can also violate federal law. It just goes to show that while AI can be incredibly powerful, it needs to be used responsibly and ethically. It's also important for regulators to do their due diligence and pay extra attention to these new applications.

San Francisco debates letting police deploy robots that kill

FILE - San Francisco Police Chief Bill Scott answers questions during a news conference in San Francisco, on May 21, 2019. The Democratic San Francisco Board of Supervisors could allow police to use potentially lethal, remote-controlled robots in emergency situations. The 11-member board will vote Tuesday, Nov. 29, 2022, on a controversial proposal opposed by civil rights advocates critical of the militarization of police. (AP Photo/Eric Risberg, File)

San Francisco has approved a plan to allow the police to use potentially lethal, remote-controlled robots in emergency situations. The decision follows an emotionally charged debate that reflected divisions among the city’s liberal-leaning supervisors over support for law enforcement. Opponents of the policy said that it would further the militarization of the police force and lead to further aggression towards poor and minority communities. The San Francisco Police Department said that it has no plans to arm robots with guns, but it could deploy robots equipped with explosive charges to contact and incapacitate suspects in situations where lives are at risk.

Our take: Thee to allow the use of potentially lethal, remote-controlled robots in emergency situations is a concerning development. It may also become a slippery slope leading to more worrying use of lethal and autonomous robots in the future. While SFPD said that it has no plans to arm robots with guns, this authorization could be expanded to include such weapons in the future. Furthermore, allowing robots to be used in life-or-death situations opens the door to the possibility of robots being used in contexts where lethal force is currently not authorized. This could pave the way for the increasing use of robots in law enforcement and the military, raising serious concerns about accountability and the potential harms.

Other News

Research

Mastering Stratego, the classic game of imperfect information - "DeepNash learns to play Stratego from scratch by combining game theory and model-free deep RL"

AI invents millions of materials that don’t yet exist - "‘Transformative tool’ is already being used in the hunt for more energy-dense electrodes for lithium-ion batteries"

New programming tool turns sketches, handwriting into code - "Cornell University researchers have created an interface that allows users to handwrite and sketch within computer code -- a challenge to conventional coding, which typically relies on typing."

Ten Years of Image Synthesis - "It’s the end of 2022. Deep learning models have become so good at generating images that, at this point, it is more than clear that they are here to stay."

New AI model calculates risk of heart attack or stroke using a single X-ray - "Researchers have developed a new artificial intelligence (AI) model capable of predicting a patient’s 10-year risk of death from a heart attack or stroke—all with a single chest X-ray. "

137 emergent abilities of large language models - "Whereas emergent prompted tasks focus on a particular dataset, the second category of emergence is few-shot prompting strategies, which are general prompting strategies that only work for language models of a sufficiently large scale."

Discover State-of-the-Art Machine Learning Demos on arXiv - " Starting today, Hugging Face Spaces is integrated with arXivLabs through a Demo tab that includes links to demos created by the community or the authors themselves. "

Pulses driven by artificial intelligence tame quantum systems - "Machine learning drives self-discovery of pulses that stabilize quantum systems in the face of environmental noise."

Applications

"The Final Resting Place of the Queen Bee" (Credit: Alexander Reben)

The weird and wonderful art created when AI and humans unite - "Will AI kill art? Not likely, says the artist Alexander Reben, who has been working with AI for years. In fact, we may be entering an exciting new period that changes how we think about creativity itself."

Artist Feeds Her Childhood Diary Into an AI Chatbot -- And Has a Moving Conversation With Her Younger Self - "What you're looking at above is an excerpt from a moving conversation between tech artist Michelle Huang, and Michelle Huang as a young girl."

A message from John Furrier, co-founder of SiliconANGLE: - "The new Create with Alexa feature on Echo Show devices relies on sophisticated conversational and generative AI models to create animated bedtime stories featuring colorful graphics, background music and a compelling narrative."

How the World Cup’s AI instant replay works - "A new hyper-accurate technology, and referees' eternal quest for objectivity."

Enter the multiverse: the chat-room game made of AI art - "An exciting multiplayer Discord game asks you to find things in the multiverse through an AI image generator. The hallucinatory results could mark a new frontier for AI art"

Disney Made a Movie Quality AI Tool That Automatically Makes Actors Look Younger (or Older) - "With just a few clicks, actors can look younger or older without the need for expensive visual effects."

Machine learning, GPS alternatives key for navigating future jammed environments - "The U.S. and its military allies rely on GPS for navigation of high-value assets, but the technology is quite vulnerable to jamming and other interference."

Business

World first: Bosch and Mercedes-Benz’s driverless parking system approved for commercial use - "Germany’s Federal Motor Transport Authority (KBA) has approved their highly automated parking system for use in the P6 parking garage run by APCOA at Stuttgart Airport."

A leaked Amazon memo may help explain why the tech giant is pushing out so many recruiters - "Amazon has quietly been developing AI software to screen job applicants."

Google Health licenses its AI breast cancer screening tool to a medical biz - "Google has licensed its AI breast cancer screening model to a commercial medical technology company, paving the way for the system developed by researchers to be tested in real clinical settings for the first time."

Amazon to warn customers on limitations of its AI - "Amazon.com Inc (AMZN.O) is planning to roll out warning cards for software sold by its cloud-computing division, in light of ongoing concern that artificially intelligent systems can discriminate against different groups, the company told Reuters."

Amazon’s Quest for the ‘Holy Grail’ of Robotics - "Automating some warehouse functions could let the giant do more with its existing workforce—or the retailer may not require all the humans it employs today"

Lensa AI climbs the App Store charts as its ‘magic avatars’ go viral - "It might seem like Lensa AI sprung up over night when suddenly, your friends are all posting artistic renditions of themselves that they generated on the app."

How tech companies are using AI to tackle clinician burnout, administrative headaches - "AI-enabled triage can shuttle patients to the ideal care provider"

Electric robot tractors powered by Nvidia AI chips are here - "Monarch Tractor, an electric smart tractor company, says its first AI-powered farming vehicles, dubbed the MK-V, are rolling off the production line."

BeeHero raises $42 million Series B for precision pollination platform - "BeeHero, which has developed a precision pollination platform, announced on Thursday that it has completed a $42 million Series B funding round."

Concerns

Now AI can write students’ essays for them, will everyone become a cheat? - "Teachers and parents can’t detect this new form of plagiarism. Tech companies could step in – if they had the will to do so"

AI experts are increasingly afraid of what they’re creating - "AI gets smarter, more capable, and more world-transforming every day. Here’s why that might not be a good thing."

Effective Altruism Is Pushing a Dangerous Brand of ‘AI Safety’ - "This philosophy—supported by tech figures like Sam Bankman-Fried—fuels the AI research agenda, creating a harmful system in the name of saving humanity"

Expert Opinions

AI Is Terrible at Detecting Misinformation. It Doesn’t Have to Be. - "A look at Elon Musk’s predicament at Twitter, and how AI might help."

AI is cognitive automation, not cognitive autonomy - "Like the rest of computer science, AI is about making computers do more, not replacing humans."

AI Reveals the Most Human Parts of Writing - "When do writers want help finding inspiration? And when do they want full control? Computers could expose the true future of the medium."

AI could have 20% chance of sentience in 10 years, says philosopher David Chalmers - "AI programs that combine sensing and action and world models might achieve fish-level intelligence, says the NeurIPS 2022 keynote speaker."

Copyright © 2022 Skynet Today, All rights reserved.

Last Week in AI

Last Week in AI #196: OpenAI's ChatGPT takes…